matlab - Filter elements from a 3D matrix without loop -

i have 3d matrix h(i,j,k) dimensions (i=1:m,j=1:n,k=1:o). use simple case m=n=o = 2:

h(:,:,1) =[1 2; 3 4]; h(:,:,2) =[5 6; 7 8];

i want filter matrix , project (m,n) matrix selecting each j in 1:n different k in 1:0.

for instance, retrieve (j,k) = {(1,2), (2,1)}, resulting in matrix g:

g = [5 2; 7 4];

this can achieved for loop:

filter = [2 1]; % meaning filter (j,k) = {(1,2), (2,1)}  = 1:length(filter)     g(:,i) = squeeze(h(:,i,filter(i))); end

but i'm wondering if possible avoid for loop via smart indexing.

you can create linear indices such output expansion needed first dimension bsxfun. implementation -

szh = size(h) offset = (filter-1)*szh(1)*szh(2) + (0:numel(filter)-1)*szh(1) out = h(bsxfun(@plus,[1:szh(1)].',offset))

how work

(filter-1)*szh(1)*szh(2) , (0:numel(filter)-1)*szh(1) gets linear indices considering third , second dimension elements respectively. adding these 2 gives offset linear indices.
add first dimenion linear indices 1:szh(1) offset array in elementwise fashion bsxfun give actual linear indices, when indexed h output.

sample run -

h(:,:,1) =      1     2      3     4 h(:,:,2) =      5     6      7     8 filter =      2     1 out =      5     2      7     4

Maxid